IEEE Transactions on Biomedical Engineering最新文献

Objective: For patients with facial palsy, the wait for return of facial function and resulting vision risk from poor eye closure, difficulty speaking and eating from flaccid oral sphincter muscles, and psychological morbidity from the inability to smile or express emotions can be devastating. There are limited methods to assess ongoing facial nerve regeneration: clinicians rely on subjective descriptions, imprecise scales, and static photographs to evaluate facial functional recovery. We propose a more precise evaluation of dynamic facial function through video-based machine learning analysis to facilitate a better understanding of the sometimes subtle onset of facial nerve recovery and improve guidance for facial reanimation surgery.

Methods: We present machine learning methods employing likelihood ratio tests, optimal transport theory, and Mahalanobis distances to: 1) assess the use of defined facial landmarks for binary classification of different facial palsy types; 2) identify regions of asymmetry and potential palsy during specific facial cues; and 3) quantify palsy severity and map it directly to widely used clinical scores, offering clinicians an objective way to assess facial nerve function.

Results: Our results demonstrate that video analysis provides a significantly more accurate and detailed assessment of facial movements than previously reported.

Conclusions: Our work allows for precise classification of facial palsy types, identification of asymmetric regions, and assessment of palsy severity.

Significance: This project enables clinicians to have more accurate and timely information to make decisions for facial reanimation surgery, which will have drastic consequences on the quality of life for affected patients.

Lung cancer is one of the leading causes of cancer-related deaths, and accurate staging is critical for determining the appropriate treatment. Robotic Navigation Bronchoscopy has shown advantages over traditional manual procedures, offering benefits in safety, efficiency, and accessibility. Although there is ongoing discussion regarding autonomous RNB, there is limited focus on the autonomy in advancing the bronchoscope. In this study, we introduce a novel method for conditional autonomy in advancing and aligning a robotic bronchoscope, which was validated in vitro, ex vivo, and in vivo. This conditional autonomy utilizes a monoscopic bronchoscopic view as input, with operators guiding the system by specifying the next airway to enter at branching points. The reachability of target lesions using this conditional autonomy was 73.3% in the phantom study and 77.5% in the ex vivo study. Statistical significance was found in success rates between bifurcations and trifurcations (p = 0.03) and across lobe segments (p = 0.005). The presence of breathing motion did not affect lesion reachability or the success of turns at branching points in the ex vivo studies. In the in vivo study, when comparing conditional automation to human-operated navigation, the conditional automation took less time to reach the target lesions than human operators. The median time for passing each bifurcation was 2.5 seconds for human operators and 1.3 seconds for conditional automation. By improving precision and consistency in tissue sampling, this technology could redefine the standard of care for lung cancer patients, leading to more accurate diagnoses and therapies.

Objective: Quantitative Susceptibility Mapping (QSM) measures magnetic susceptibility of tissues, aiding in the detection of pathologies like traumatic brain injury, cerebral microbleeds, Parkinson's disease, and multiple sclerosis, through analysis of variations in substances such as iron and calcium. Despite its clinical value, using high-resolution QSM (voxel sizes < 1 mm³) reduces signal-to-noise ratio (SNR), which compromises diagnostic quality.

Methods: Denoising of T₂^*-weighted (T₂^*w) data was implemented using Marchenko-Pastur Principal Component Analysis (MP-PCA), allowing to enhance the quality of R₂^*, T₂^*, and QSM maps. Proof of concept of the denoising technique was demonstrated on a numerical phantom, healthy subjects, and patients with brain metastases and sickle cell anemia.

Results: Effective and robust denoising was observed across different scan settings, offering higher SNR and improved accuracy. Noise propagation was analyzed between T₂^*w, R₂^*, and T₂^* values, revealing augmentation of noise in T₂^*w compared to R₂^* values.

Conclusions: The use of MP-PCA denoising allows the collection of high resolution (∼0.5 mm³) QSM data at clinical scan times, without compromising SNR.

Significance: The presented pipeline could enhance the diagnosis of various neurological diseases by providing higher-definition mapping of small vessels and of variations in iron or calcium.

The open-source development model has been successfully applied to consumer and enterprise software, and recently to consumer hardware. Medical devices may become a beneficiary of this trend, as open-source medical device development has the potential to reduce costs, democratize patient access, and provide continued support to abandoned devices from failed companies. Unlike the consumer device market, the medical device market is highly regulated and involves considerable manufacturer liability that may limit the use of open-source technology. This review of open-source medical device development explores the current state of development in research and clinical products and suggests best practices for creating sustainable and effective open-source medical devices.

A transfer function approach has recently proven effective for calibrating deep learning (DL) algorithms in quantitative ultrasound (QUS), addressing data shifts at both the acquisition and machine levels. Expanding on this approach, we develop a strategy to acquire the functionality of a DL model from one ultrasound machine and implement it on another in a black-box setting, in the context of QUS. This demonstrates the ease with which the functionality of a DL model can be transferred between machines. While the proposed approach can also assist regulatory bodies in comparing and approving DL models, it also highlights the security risks associated with deploying such models in a commercial scanner for clinical use. The method is a black-box unsupervised domain adaptation technique that integrates the transfer function approach with an iterative schema. It does not utilize any information related to model internals but it solely relies on the availability of an input-output interface. Additionally, we assume the availability of unlabeled data from a testing machine. This scenario could become relevant as companies begin deploying their DL functionalities for clinical use. In the experiments, we used a SonixOne and a Verasonics machine. The model was trained on SonixOne data, and its functionality was then transferred to the Verasonics machine. The proposed method successfully transferred the functionality to the Verasonics machine, achieving a remarkable 98 percent classification accuracy in a binary decision task. This study underscores the need to establish security measures prior to deploying DL models in clinical settings.

Although terahertz (THz) metasurfaces based on bound states in the continuum (BIC) have garnered significant attention in biomedical applications, their technical implementation in high-sensitivity cancer cells detection remains a critical challenge. In this work, we present a THz biosensor employing dual split-ring resonator (DSRR) arrays quasi-bound states in the continuum (Q-BIC). Numerical simulations reveal a high-Q resonance dip at 2.35 THz with a detection sensitivity of 522 GHz/RIU. Experimentally, the performance was validated by detecting normal cells (murine splenocytes) and three cancer cell lines (LLC, LoVo, and MC38). In addition, analysis of cell type discrimination was achieved by integrating machine learning algorithms to project high dimensional spectral data into a low-dimensional space. This study establishes a label-free approach for long-term cellular monitoring, advancing THz technology as an innovative platform for practical biomedical applications.

In the intensive care unit (ICU), monitoring sedation levels is crucial. Clinicians often rely on intermittent behavioral scales like the Richmond Agitation-Sedation Scale (RASS), which can be subjective and delay timely interventions. While electroencephalography (EEG) offers a continuous and non-invasive alternative, but the complexity of consciousness renders a unimodal signal insufficient for comprehensive representation. To address these challenges, we propose a novel multimodal deep learning framework, Hierarchical Multimodal Fusion with Dynamic Correction (HMDC), that synergistically integrates EEG with peripheral physiological signals including blood pressure, heart rate, and oxygen saturation. The architecture features a dual-stream pathway to process both raw temporal EEG data and its spectral features from spectrograms. These neural representations are then intelligently fused and refined by a Dynamic Correction Module using a confidence-weighting mechanism. The model was developed and validated on a dataset comprising 2,880 labeled RASS assessments from 105 ICU patients, with scores ranging from -5 (comatose) to +1 (restless). The HMDC framework achieved a classification accuracy of 83.8%, significantly outperforming unimodal and simpler fusion baselines. By providing a temporally precise and physiologically grounded sedation assessment, this integrative approach establishes a robust correlation between multimodal signal patterns and clinical states, offering clinicians a unified tool for optimizing sedative titration and potentially minimizing delirium risks.

Objective: Photoplethysmography (PPG) is widely used for cardiovascular monitoring, but its analysis is challenged by signal variability, inconsistent acquisition settings, and limited interpretability. This study investigates the use of low-dimensional embeddings to support down-stream tasks, including anomaly detection, activity classification, and signal authenticity verification across diverse PPG modalities.

Methods: We developed a pipeline lever aging dimensionality reduction techniques, Autoencoder (AE), Fully Connected Neural Network (FCNN), and Uniform Manifold Approximation and Projection (UMAP) to extract compact signal representations. These methods were evaluated across four datasets representing clinical (BIDMC, MIMIC-PERFORM), wearable (Wrist PPG), and remote PPG (UBFC) recordings. Performance was assessed through clustering indices, classification metrics, and anomaly detection rates under varying noise levels.

Results: Quantitative evaluation demonstrated that AE-based embeddings enabled accurate discrimination between neonatal and adult signals in the MIMIC-PERFORM dataset (F1 = 0.92, AUC = 0.90), while UMAP outperformed AE and FCNN in clustering physical activities from Wrist PPG data (Davies Bouldin Index = 5.40). In the BIDMC dataset, the framework detected synthetic anomalies with an AUC of 0.77 at 2 dB SNR, with detection rates declining consistently with reduced noise. On the UBFC dataset, UMAP embeddings supported the detection of manipulated rPPG signals with an F1 score of 0.75 and an AUC of 0.73.

Conclusion: Low dimensional representations provide a compact and task relevant encoding of PPG signals that enhances classification and detection performance in multiple scenarios. While interpretability gains remain task-dependent, these findings support the utility of embedding-based approaches in biomedical signal analysis and their robustness across modalities and noise conditions.

Objective: Tumor Treating Fields (TTFields) therapy, a clinically established modality that disrupts cancer cell mitosis through biophysical mechanisms, presents a unique paradigm in oncology. Despite its proven efficacy, its broad application is hindered by significant challenges in optimizing treatment delivery for individual patients.

Methods: This review synthesizes the landscape of advanced computational strategies designed to overcome these barriers. We argue that personalizing TTFields therapy requires tackling three interdependent obstacles: achieving accurate electric field dosimetry, ensuring thermal safety, and enabling adaptive treatment planning.

Results: this review systematically analyzes the state-of-the-art computational solutions corresponding to each challenge. We first examine patient-specific electric field modeling, emphasizing the critical roles of high-fidelity segmentation and quantitative dosimetric criteria. We then delve into thermal safety analysis, focusing on coupled electro-thermal simulations for predicting and mitigating thermal risks. Finally, we explore the multifaceted approaches to personalization, reviewing the convergence of algorithmic array layout optimization, real-time monitoring systems, and synergistic surgical interventions.

Significance: By structuring the current body of research within this "problem-solution" framework, this review provides a clear and cohesive synthesis of how computational engineering is paving the way for a new era of precise, safe, and adaptive TTFields therapy.

Objective: Speech, as the core of advanced human cognition, is fundamental to social interaction and daily life. Electroencephalogram (EEG)-based speech brain-computer interface (BCI) offers a novel communication pathway for patients with speech disorders, where deep learning has demonstrated significant advantages. Given the established dominance of the left hemisphere in speech processing, exploring methods to extract speech related neural features fully is crucial for enhancing decoding per formance.

Approach: In this study, EEG signals were recorded during a silent speech task involving the articulation of 10 distinct Chinese characters. Leveraging the principle of language function lateralization, we proposed a novel deep learning model, the cross hemispheric spatial-temporal attention network (CHSTAN), for EEG-based silent speech recognition. A multiscale temporal con volution block was employed to extract the temporal dynamics of EEG signals. A hemispheric spatial convolutional block was designed to independently process spatial information from the left and right hemispheres. Furthermore, the cross-attention mechanism was introduced to enhance inter-hemispheric feature inter action and specifically reinforce left-hemispheric feature representation for the final classification.

Results: We compared CHSTAN with other existing methods using 5-fold cross-validation on the collected dataset. CHSTAN achieved an average classification accuracy of 49.88% and an average F1-score of 48.75% in decoding the 10 Chinese characters, significantly outperforming other methods.

Conclusion: The results indicate that the CHSTAN performs effectively in silent speech EEG classification tasks. Notably, the feature patterns learned through its innovative architecture correspond to neural speech processing mechanism.

Significance: CHSTAN provides valuable insights and practical solutions for improving the performance of EEG-based speech decoding.